Content & CreativeDone with Claude Code Mar 26, 2026
Analyze a piano recording's BPM, structure and stems
With a skill installed, Claude Code installed ffmpeg and librosa and ran a Python pipeline on a local MP3, producing a report on BPM, structure and stems.
- What They Did
- The author had Gemini draft a skill on Loooom, then published it through the web UI and installed it in Claude Code. He asked Claude to identify parts of a song and pointed it at a local MP3 of his 37-second piano clip. Claude installed ffmpeg and librosa mid-session and ran a Python pipeline.
- How It Went
- It produced a report with BPM, meter, key, a four-section timeline and a list of stems. The BPM was estimated at about 91 against a canonical ~104. The stem list included vocals and drums, though the clip was solo piano.
- Worth Knowing
- Expect rough accuracy: the BPM was off from the real tempo, and the stems listed (kick, vocal, hi-hat) don't match a solo piano recording.
Try it yourself with Claude Code
I have a piano recording at [path to MP3 file]. Install whatever you need (such as ffmpeg and librosa), then analyze it and write me a report covering the BPM, the song structure with approximate timestamps for each section, and separated stems if possible. Save the report as [report filename] in the same folder and tell me when it's finished and where the output files are.